Machine Learning algorithm such as Support Vector Machine (/SVM) are identified to be significant for prediting the variables related to the pre-defined output in real world applications. This machine learning model when imposed over the encrypted data is highly indispensable for protecting data and model information against malicious attackers. These malicious adversaries launch the attack over the data either during the phase of prediction or training. Torus-based Fast Fully Homomorphic Encryption (TFFHE) scheme is identified to facilitate potential evaluations of encrypted data related to real numbers, this merits of this TFFHE motivated the option of implementing a privacy preserving machining algorithm which can be utilized during the process of training. In this paper, Secure SVM Training using Privacy Preserving Homomorphic Encryption Algorithm using TFFHE is proposed for preventing inefficient operations and numeric stability during the phase of training in the encrypted domain. This TFFHE is proposed based on the improvement of GSW and its associated ring variants. With respect to real world datasets, this TFFHE-based SVM confirmed better performance on par with the state of the art FHE and logistic regression classifiers used for comparison. This study to the best of the knowledge is one of the few practical algorithms which could be used for SVM model training with the integration of FHE.